Part correlation in JASP
I see under Correlation that JASP can report partial correlation (and p-value thereof). Bit of a shame that it also can't report semi-partial correlation (part correlation).
But I've found a way in JASP to get this: do a linear regression and request "part and partial correlations". But it doesn't report p-values.
Not a problem for partial correlation as can get that p-value under aforementioned partial correlation route. But it is an extant problem for part correlation. Request that JASP can add this. Plus ask how I might get a p-value in the meantime?
Thank you.
Comments
I believe the p-value will match exactly that of the regression coefficient, as they are "semi-partial" effects themselves (i.e. unique variance only)
The p-value of the regression coefficient appears to be the same as that for partial correlation.
How can I get p-value for semi-partial correlation (also called part correlation)?
Thank you.
Having to do this outside of JASP: does the following look ok? (and by all means please think about incorporating this into JASP, or some fix of it, if you think the following is wrong)
JASP doesn’t report p-values for semi-partial correlations. To calculate such p-values the following lines of R code (following [kim1, kim2]) were used, where N is the sample size, g is the number of controlled variables (which is 1 here), R is the semi-partial correlation: df=N-2-g. Tog=R*sqrt(df/(1-R^2)). pt(Tog, df, lower = F)*2.
[kim1] Kim S (2015) ppcor: an R package for a fast calculation to semi-partial correlation coefficients. Communications for statistical applications and methods. 22(6):665.
[kim2] Kim S (2022) P-value calculation methods for semi-partial correlation coefficients. Communications for statistical applications and methods. 29(3):397.
It would be a good feature request for our GitHub page!
(for details see https://jasp-stats.org/2018/03/29/request-feature-report-bug-jasp/)
EJ
BTW you can compute the semi-partial r from the standard r's, as outlined in here for instance:
https://personal.utdallas.edu/~herve/Abdi-PartialRegressionCoefficient2007-pretty.pdf
Under linear regression, JASP reports semipartial correlations (part correlations) but NO corresponding p-value. To repeat for emphasis, I am talking about semipartial correlation here.
Might this get the p-value:
Put the other predictor(s) into the null model. Then the p-value shown for the R^2 change is that for the semipartial correlation?
Becausethe semipartial correlation is the square root of the R^2 change.
Is this valid? Or nonsense?
Hi Both,
Just a quick note to say someone made a feature request for this a few months ago
[Feature Request]: add 'semipartial correlation' analysis under regression · Issue #2244 · jasp-stats/jasp-issues (github.com)